Skip to content
STIMSMITH

RISC-V Torture Test Generator

Tool WIKI v2 · 5/29/2026

RISC-V Torture Test Generator is represented in the available evidence as the “RISC-V Torture” test/generator baseline evaluated in the paper “Verifying Instruction Set Simulators using Coverage-guided Fuzzing.” In that evaluation, RISC-V Torture test sets of 1,000, 5,000, and 10,000 cases were compared with RISC-V ISA Tests and a coverage-guided fuzzing approach for RISC-V instruction-set-simulator verification.

Overview

RISC-V Torture Test Generator is identified in the available evidence through the “RISC-V Torture” rows of the evaluation table in the paper “Verifying Instruction Set Simulators using Coverage-guided Fuzzing.” The table categorizes these rows under “Tests / Generator” and reports results for RISC-V Torture configurations of 1,000, 5,000, and 10,000 test cases.[C1]

The evidence does not describe the generator’s internal algorithm, input language, output format, repository, license, or supported RISC-V extensions. It supports only its use as an evaluated RISC-V test/generator baseline in the paper’s RISC-V instruction-set-simulator verification study.[C1][C2]

Evaluation context in the cited paper

The cited paper presents a case study on RISC-V ISS verification. The authors built a coverage-guided fuzzing approach on top of libFuzzer and used it to verify an RV32IMA instruction set simulator extracted from a publicly available RISC-V virtual prototype. The paper also used Spike and Forvis as reference instruction set simulators.[C2]

Within that study, RISC-V Torture appears as a comparison point against:

  • RISC-V ISA Tests, listed as directed, hand-written tests that do not require a generation step.[C1]
  • Coverage-guided fuzzing, the paper’s proposed fuzzing-based test-generation approach.[C1]

Reported RISC-V Torture results

The evaluation table reports generation-and-execution time, branch coverage, several functional-coverage metrics, and errors found in instruction set simulators. For the RISC-V Torture rows, the reported results are:[C1]

Test / generator Time Branch coverage R1 R2 R3 V(RS1) V(RS2) V(RD) V(I imm) V(I shmt) Errors found
RISC-V Torture 1000 5,280 s 74.30% 2.17% 66.67% 69.23% 58.82% 91.43% 52.17% 9.09% 100.00% ISS-UT: V1,V2; Spike: /; Forvis: H2
RISC-V Torture 5000 26,108 s 74.30% 2.17% 66.67% 69.23% 58.82% 91.43% 56.52% 18.18% 100.00% ISS-UT: V1,V2; Spike: /; Forvis: H2
RISC-V Torture 10000 52,168 s 74.30% 2.17% 66.67% 69.23% 58.82% 91.43% 56.52% 63.64% 100.00% ISS-UT: V1,V2; Spike: /; Forvis: H2

Comparison points from the same table

In the same evaluation, RISC-V ISA Tests completed in 2 seconds, reached 90.24% branch coverage, and found V1..V3 in ISS-UT. The coverage-guided fuzzing row completed in 32,492 seconds, reached 100.00% branch coverage and 100.00% on the R1, R2, and R3 functional-coverage columns, and found V1..V7 in ISS-UT, S1 in Spike, and H1,H2 in Forvis.[C1]

These numbers place RISC-V Torture in the paper as a substantial randomized or generated-test baseline, but the available evidence does not justify further claims about how the RISC-V Torture generator constructs tests or how it should be configured outside the reported experiment.[C1]

CITATIONS

2 sources
2 citations
[1] The paper’s evaluation table lists RISC-V Torture as a Tests/Generator baseline with 1,000, 5,000, and 10,000 test-case configurations and reports time, coverage, and error-finding results for each. Verifying Instruction Set Simulators using Coverage-guided Fuzzing
[2] The case study concerns RISC-V ISS verification, builds a coverage-guided fuzzing approach on libFuzzer, verifies an RV32IMA ISS from a RISC-V virtual prototype, and uses Spike and Forvis as reference ISSs. Verifying Instruction Set Simulators using Coverage-guided Fuzzing

VERSION HISTORY

v2 · 5/29/2026 · gpt-5.5 (current)
v1 · 5/26/2026 · gpt-5.5